Troubles with ggplot and geom_bar - r

Question

Here the updated example:
df <- data.frame(a=rep(c("A","B"),each=10),
b=rep(rep(c("C","D"),each=5),2),
c=c(sample(letters[1:5]), sample(letters[6:10]),
sample(letters[1:5]), sample(letters[6:10])),
d=c(0.10,0.18,0.34,0.35,0.59,0.16,0.38,0.40,0.53,0.58,
0.37,0.62,0.83,1.46,-0.91,-0.79,-0.52,-0.43,-0.01,0.34))
> df
a b c d
1 A C b 0.10
2 A C e 0.18
3 A C a 0.34
4 A C c 0.35
5 A C d 0.59
6 A D i 0.16
7 A D j 0.38
8 A D h 0.40
9 A D f 0.53
10 A D g 0.58
11 B C e 0.37
12 B C d 0.62
13 B C a 0.83
14 B C c 1.46
15 B C b -0.91
16 B D f -0.79
17 B D i -0.52
18 B D h -0.43
19 B D j -0.01
20 B D g 0.34
If you look closely, you will see that column d is ordered within column b always from smallest to largest.
The first plot is how I would like to have the plot apart from the fact, that the bars displayed are not in the order of d. So the bars do not appear from smallest to largest:
p <- ggplot(df, aes(x=c, y=d, fill=b, stat="identity")) +
facet_grid(. ~ a) +
geom_bar()
print(p)
This is because column c is a factor and the factors are apparently not ordered in the same order as column d is. So I did the following:
df$c <- paste(1:nrow(df), df$c, sep="_")
df$c <- factor(df$c, levels = unfactor(df$c))
p <- ggplot(df, aes(x=c, y=d, fill=b, stat="identity")) +
facet_grid(. ~ a) +
geom_bar()
print(p)
produces the following plot:
Here the order is correct. However, as you can see I created unique factors I get those spaces for the ones not present in A and B respectively.
How can I sort that out?

Answer 1

Now that you have changed the question, 'ggplot' cannot do this for you. By giving [df$c] levels, you could order the data but only based on the first set of [c] values. For instance:
df$c <- factor(df$c, levels=levels(df$c)[order(df$d)])
But that won't work, since you're trying to sort [df$c] twice (once for "A", and once for "B").
You really need to break this into two separate plots, and just plot the two viewports next to each other.
Setting up the viewports:
grid.newpage()
pushViewport(viewport(layout = grid.layout(1, 2)))
Plot A:
a_df <- df[df$a=="A",]
a_df$c <- factor(a_df$c, levels=levels(a_df$c)[order(a_df$d)])
a_p <- ggplot(a_df, aes(x=1:10, y=d, fill=b)) +
facet_grid(. ~ a) +
geom_bar(stat="identity", position="dodge")
print(a_p, vp = viewport(layout.pos.row=1, layout.pos.col=1))
Plot B:
b_df <- df[df$a=="B",]
b_df$c <- factor(b_df$c, levels=levels(b_df$c)[order(b_df$d)])
b_p <- ggplot(b_df, aes(x=1:10, y=d, fill=b)) +
facet_grid(. ~ a) +
geom_bar(stat="identity", position="dodge")
print(b_p, vp = viewport(layout.pos.row=1, layout.pos.col=2))
From here, you can worry about removing the excess legend, choosing which axes to label and such, but it looks exactly like your example plot only with the empty locations removed.
This is really an example of how 'ggplot' is sometimes more of a hindrance and less of a boon. In my experience, it's best to first design your plot and then choose the tool. Frequently, I find myself going back to raw 'grid' to do my visuals, because I want something the 'grid' wrapper 'ggplot' just won't do.
Note: In the future, don't delete your original question content; just add the updated info. Removing the old content makes a lot of the answers and comments on this page irrelevant.

Answer 2

I think this is actually a common mistake with the 'ggplot' function. If you set an outline color (i.e. aes(colour="red")), you will see that you are actually plotting all four values, but they are plotting on top of each other. The stacking warning is because the default value of 'position' is "stack". Just include the position="dodge" argument, and that will go away.
Now, to actually solve your problem. You need to give 'ggplot' something to distinguish between the values of X(A), X(B), Y(A), and Y(B). At first glance, you might be tempted to use your [b] values, but you don't want all of the extra spaces. Let's adjust your dataframe to have only 1s and 2s for [b]:
df <- data.frame(a=rep(rep(c("A","B"),each=2),2),
b=rep(1:2,4),
c=rep(c("X","Y"),each=4),
d=c(1.2,1.1,1.15,1.1, -1.1,-1.05,-1.2,-1.08))
The plot is actually pretty easy to fix once you know the problem. First, set [b] to your x-axis, and add [a] to your facet. Then remove all of the annoying gibberish from [b] using the 'theme' with blank elements:
p <- ggplot(NULL, aes(x=b, y=d)) +
facet_grid(. ~ c + a) +
geom_bar(data = df, stat="identity", position="dodge") +
theme(axis.ticks = element_blank(), axis.text.x = element_blank(), axis.title.x = element_blank())
print(p)
If this isn't exactly what you want, it should be at least close enough that you'll only have to do cosmetic changes. Good luck!